Catch application runtime bugs in an error boundary and record stacktraces in telemetry

[ericpgreen2]

For capturing runtime errors:

• Svelte does not have a native error boundary. See: Error Boundary RFC sveltejs/rfcs#46.
• The oft-cited third-party Svelte library is the Crown Framework, but it is unmaintained.
• Hooking into window.onerror and window.onunhandledrejection seems like one valid option.

window.onerror = (msg, url, line, col, error) => {
    // capture error & emit telemetry
    // show a full error page
};
window.onunhandledrejection = (event) => {
    // capture error & emit telemetry
    // show a full error page
};

• Sentry implies their Svelte SDK captures runtime errors

For recording stacktraces in telemetry:

• Let's use Sentry
• Ideally we proxy the requests through our backend

[AdityaHegde]

Done as part of #3516

[ericpgreen2]

Re-opening because we still need to do this for Rill Developer.

[ericpgreen2]

Notes for the second pass:

• Telemetry for Rill Developer (as mentioned above)
• Capture more information for the "unknown error" case. It's happening frequently. Here's some code from ChatGPT that might be helpful:

  if (!error.response) {
    if (error.code === 'ECONNABORTED') {
      console.log('Request timed out');
    } else if (error.message.includes('Network Error')) {
      console.log('Network error - the request was made but no response was received');
    } else if (error.message === 'Request aborted') {
      console.log('Request was aborted');
    } else {
      console.log('Unknown error occurred');
    }
  }

• Capture the full page route. This will inform us of the environment dev/test/stage/prod, the specific dashboard, and any query parameters.
• Capture the Rill application version

[AdityaHegde]

There were some challenges to get this working locally and not have a lot of spam. Especially when actively modelling there are a lot of errors for the resource API when the model is not valid.